building AI systems that work beyond the demo. agents, automation & the thinking behind them.
San Francisco, CA
Based in Côte d'Ivoire
- Tweets 2,146
- Following 512
- Followers 486
- Likes 5,294
You slept 8 hours. Your AI team didn't.
The shift from an assistant you watch to agents that run while you're offline: the one rule that keeps them safe, the two triggers that wake them, and 7 job specs you can paste today.
Bookmark it.
155
99
35,626
capt retweeted
buying into $FET here - clean breakout and retest ongoing
tenkan 3d, kijun 1d, cloud support all at 20c
i prefer 3d for htf which frontruns weekly (which just flashed early buy on superkumo)
dcaing down to 20c, pretty good setup
long term spot plan below
30
106
1,068
101,352
Wizz season $FET 👀✍️
36
26
119
Personal AI should not be limited to one device.
ASI:One brings the same intelligent interface to mobile.
Plan. Continue workflows. Manage agentic bookings. Do it wherever the task happens.
Get the app below 👇
App Store: apps.apple.com/us/app/asi-on…
Google Play: play.google.com/store/apps/d…
11
51
413
19,098
capt retweeted
Your fav influencers all watching the wizz timeline
Study
$NEAR $3 + soon.
34
53
779
95,514
$ZEN is trading at $6.78.
Meanwhile, this thing has traded above $100 before.
That gap alone is why I keep watching it.
Right now $ZEN is sitting near the bottom of a structure
that has been building for years.
If this liquidity zone holds and we finally get a proper expansion
there’s a ridiculous amount of empty space above.
My targets:
$46 → $138 → $169
And if $ZEN enters a real mania phase, I’m watching $220+.
That’s the scenario on my chart that offers roughly +5,400% from the lows.
The funny thing about moves like this?
At $6 everyone calls the target delusional.
At $100 they start asking where to buy.
7
14
71
7,011
capt retweeted
$ZEN is moving almost at $8
Last time zcash had a run towards 750 this one ran to $25
I think it will be over $10 soon
24
31
365
76,911
Claude Fable 5.1: 55.8% on Terminal-Bench 4.0.
Opus 5: 52.3%. Old Fable 5: 42.0%.
A ".1" release just jumped past Opus 5 on agentic coding. And it runs cheaper.
The quieter number is the bigger one. On Terminal-Bench-Science 0.1, Fable 5.1 scores 52.6%. Fable 5 scored 24.7%. It more than doubled its own science score in a single point release.
Same token price as Fable 5. $10 in, $50 out. Nothing moved on the sticker.
What moved is effort. It reaches the same answer at a lower effort level, so typical jobs cost about 25% less to run, and heavy agent jobs up to ~45% less. Anthropic's own line: "similar or better results at a much lower cost."
Proof it is not a benchmark mirage: a builder had Fable 5.1 one-shot a playable Mario Kart game. Not a snippet. The whole thing, first prompt.
If you run agents that ship real work, this is a free upgrade you have to opt into. Repoint your most expensive task at Fable 5.1 and measure cost per finished task, not per token.
The frontier didn't post a shiny new headline number this week. It got cheaper to reach.
Bookmark this before your next model pick.
We’re introducing Claude Fable 5.1 and Claude Mythos 5.1.
They're the world’s most advanced models for coding and knowledge work.
4
157
96
13,773
Receipts, so you can verify before you repoint anything.
Terminal-Bench 4.0, Terminal-Bench-Science 0.1 and the pricing are all in Anthropic's own post: anthropic.com/claude-fable-a…
Builders: what is the first task you are moving to Fable 5.1?
1
125
71
944
capt retweeted
Stanford, November 1, 2024
A mathematician stood in front of a room full of PhDs and said the one sentence every quant fund hopes he never says out loud.
"The computer lies. Thousands of hours of simulations give persistently wrong results."
His name is Persi Diaconis. He's spent decades studying how fast simulations actually reach the truth. You've probably fallen for this trap yourself without knowing it had a name. Run a backtest a thousand times, watch it come back stable every single time, and that consistency starts to feel like proof.
It isn't.
Here's his key finding. Some simulations don't take twice as long to reach the real answer. Not ten times longer. Sometimes ten thousand times longer than you'd expect. And the whole time it hasn't converged yet, it still looks stable. It looks correct. There's no way to tell the difference just by watching the output.
That's the entire trap.
Every backtest is a simulation. Every risk model, every Monte Carlo run, every machine learning training loop. Running an AI to test a thousand backtests a day doesn't fix this. It just runs the same wrong simulation a thousand times faster, and hands you a thousand confident wrong answers instead of one.
His answer isn't more computing power. It's proof, math that doesn't depend on how many times you ran it. He says most researchers kept the wrong half of the lesson. They remembered proof is hard. They forgot why it matters.
Renaissance Technologies, the highest-returning fund in history, leans on proof before anything gets deployed. Its founder was a mathematician first, a trader second.
Ask yourself this about your own backtest tonight. Did it actually converge, or did it just run long enough to look like it did?
The lecture is on YouTube. It has almost no views.
14
10
72
4,610
SpaceXAI engineer (ex-Cursor):
"right now I'm running 10-20 GrokBot agents that automate 90% of my routine
i have a Chief of Staff agent. He knows about all my other bots and manages everything"
in a 50-minutes podcast, a SpaceXAI engineer showed how to build a team of agents that will work for you 24/7
worth more than a $500 course on agentic engineering
watch today, then read how to build a Grok agents team from scratch in the article below
17
31
239
45,062
SpaceXAI engineer, Lauren Tan:
"GrokBot is the most powerful agentic tool we have ever built, but only 1% of users use it correctly
right now I'm running a team of 20+ GrokBot agents. I have a Chief of Staff agent, 3 managers and 16 workers - that's how the team looks like"
In a 1-hour workshop, a SpaceXAI engineer reveals how to use Grok agents at 100% of their potential
worth more than a $500 agent engineering course on the internet
Skip Netflix and watch today, it will change the way you use GrokBot forever, then read the article below
143
674
5,831
908,666